xlstream-eval: wildcard criteria match text cells only

[cilladev]

Fixes: #142

Problem

=SUMIF(range, "*", sum_range) matched every cell type, including numbers. The wildcard arm of Criteria::matches coerced each value to text before matching, so "*" (and any wildcard) matched numbers. Excel only matches text cells with wildcards.

Fix

The Criteria::Wildcard arm now matches Value::Text only. Numbers, booleans, dates, errors, and blanks never match a wildcard, regardless of their text rendering — matching Excel.

Changes

• crates/xlstream-eval/src/criteria.rs: wildcard arm matches Value::Text only; drop now-unused coerce import.
• Two unit tests (wildcard_skips_non_text, wildcard_does_not_match_numeric_pattern) — both fail pre-fix, pass post-fix.
• Conformance fixture issue-142-sumif-wildcard-text-only.xlsx (Excel-populated cached values) covering "*", "y*", "?es", plus controls ("<>" NonBlank and ">100") proving non-wildcard criteria still include numbers.

Scope note — separate pre-existing bug (not fixed here)

While verifying, I found COUNTIF/COUNTIFS and AVERAGEIF with wildcard/operator criteria return wrong values: try_operator_criteria re-folds already-finished per-bucket aggregates, which is correct for SUM/MIN/MAX/PRODUCT but counts buckets (not rows) for COUNT and averages-of-averages for AVERAGE. This is independent of wildcards and of this fix, and is already tracked in #174. A correct fix is architectural (store deferred FoldStates instead of finished Values), so it is intentionally out of scope. The fixture is SUMIF-only to avoid entangling #174.

Test plan

• Regression unit tests added and pass
• Conformance fixture created (Excel-populated) and passes
• cargo test -p xlstream-eval passes (2261 lib + 137 doc, no regressions)
• make check passes

Closes #142